An Intelligent LLM-Based Mental Coping Detection Framework

Social media-based Mental Health (MH) informatics often focuses on disorder detection, overlooking psychological aspects like coping mechanisms and stressors. Furthermore, user-generated texts, although contains MH signals at scale, lack clinical validation. This paper presents a two-stage framework for mental coping detection that combines a coping strategy classifier and a Retrieval-Augmented Generation (RAG) system. The coping classifier identifies coping strategies from user text, while the RAG generates responses grounded in expert-validated knowledge bases and the coping labels. A pilot study demonstrates that the classifier achieves 79.7% accuracy and 0.797 weighted F1 on a held-out dataset. The RAG system produces structured, readable, and expert-verified responses. The classifier-conditioned RAG architecture provides responses grounded in validated psychological knowledge, offering a scalable approach to bridging social media data with evidence-based insights.